def get_net_data(BEG, END):
beg_idx, end_idx = env.get_data_idxs_range(BEG, END)
raw_dates = env.get_raw_dates(beg_idx, end_idx)
input = env.get_input(beg_idx, end_idx)
px = env.get_adj_close_px(beg_idx, end_idx)
px_pred_hor = env.get_adj_close_px(
beg_idx + get_config().PRED_HORIZON,
end_idx + get_config().PRED_HORIZON)
tradeable_mask = env.get_tradeable_mask(beg_idx, end_idx)
port_mask = env.get_portfolio_mask(beg_idx, end_idx)
ds_sz = px_pred_hor.shape[1]
raw_dates = raw_dates[:ds_sz]
raw_week_days = np.full(raw_dates.shape, 0, dtype=np.int32)
for i in range(raw_dates.shape[0]):
date = date_from_timestamp(raw_dates[i])
raw_week_days[i] = date.isoweekday()
input = input[:, :ds_sz, :]
tradeable_mask = tradeable_mask[:, :ds_sz]
port_mask = port_mask[:, :ds_sz]
px = px[:, :ds_sz]
labels = (px_pred_hor - px) / px
batch_num = get_batches_num(ds_sz, get_config().BPTT_STEPS)
return beg_idx, ds_sz, batch_num, raw_dates, raw_week_days, tradeable_mask, port_mask, px, input, labels
def get_net_data(env, BEG, END):
beg_idx, end_idx = env.get_data_idxs_range(BEG, END)
raw_dates = env.get_raw_dates(beg_idx, end_idx)
input = env.get_input(beg_idx, end_idx)
px = env.get_adj_close_px(beg_idx, end_idx)
tradeable_mask = env.get_tradeable_mask(beg_idx, end_idx)
raw_week_days = np.full(raw_dates.shape, 0, dtype=np.int32)
for i in range(raw_dates.shape[0]):
date = date_from_timestamp(raw_dates[i])
raw_week_days[i] = date.isoweekday()
return beg_idx, end_idx, raw_dates, raw_week_days, tradeable_mask, px, input
def build_time_axis(raw_dates):
dt = []
for raw_dt in np.nditer(raw_dates):
dt.append(date_from_timestamp(raw_dt))
return dt
prediction = np.zeros((total_stocks))
fri_px = None
mon_px = None
fri_tradable_mask = None
mon_tradable_mask = None
cash = 1
pos = np.zeros((total_stocks))
pos_px = np.zeros((total_stocks))
eq = np.zeros((trading_days))
long_pos_mask = np.full((total_stocks), False)
short_pos_mask = np.full((total_stocks), False)
for idx in range(trading_days):
date = date_from_timestamp(raw_dates[idx])
curr_px = adj_px[:, idx]
_tradable_maks = tradable_maks[:, idx]
open_pos = date.isoweekday() == 1 and fri_px is not None
# open_pos = date.isoweekday() == 5
close_pos = date.isoweekday() == 5
predict = date.isoweekday() == 5
if predict:
fri_px = curr_px
fri_tradable_mask = _tradable_maks
selection = pred_df['date'] == date
curr_pred_df = pred_df.loc[selection]
prediction[curr_pred_df['stk_idx'].